def transform_q2s(question):
#### Input: question $q$,
# # lower case everything (why?)
# question = question.lower()
#### dependency parsed.
q = Sentence(get_parse(question), question, "en", print_tokens="new")
doc = nlp(question)
lemmas = {token.text.lower(): token.lemma_ for token in doc}
past_tense = {token.text: token.tag_ == "VBD" for token in doc}
# if question doesn't start with "why", stop there
if q.word(1).lower() != "why":
return ("ERROR: can't transform: does not start with 'why'")
# replace "'s" with "is" and "'d" with "did" where appropriate
question, q = replacements(q, {"old": "'s", "new": "is", "pos": "V"})
question, q = replacements(q, {"old": "'re", "new": "are", "pos": "V"})
question, q = replacements(q, {"old": "'d", "new": "did", "pos": "M"})
question, q = replacements(q, {"old": "'ve", "new": "have", "pos": "V"})
# remove the final question mark
if q.word(len(q.tokens)) == "?":
q.cut(len(q.tokens))
# if question starts with "why is it that", return the rest
if (q.words(1, 5).lower() == "why is it that ") or (q.words(1, 5).lower()
== "why was it that "):
for i in range(4):
q.cut(1)
return cleanup_statement(q)
#### Start at the root of $q$
all_indices = [t["index"] for t in q.tokens]
root_index = [i for i in all_indices if "ROOT" in q.find_dep_types(i)][0]
#### $subj$ = \textsc{nsubj} or \textsc{nsubjpass} %dependent of the root
subj_end_index, subj_is_plural = get_subj(q, root_index, lemmas)
#### $vp^{(\text{lemma})}$ = all remaining dependents
# (everything after subj_end_index is the lemmatized vp)
#### if $aux$ in [``do'', ``does'', ``did'']
if (q.word(2).lower() in ["do", "does", "did"]):
#### $vp$ = apply tense/person of $aux$ to $vp^{(\text{lemma})}$
q = conjugate_main_verb(q, q.word(2), root_index)
else:
#### else $vp$ = $aux$ $vp^{(\text{lemma})}$
# if it's not "do", just move the single helping word to after the subj NP
q.move(2, subj_end_index)
#### $s$ = $subj$ $vp$
#### Remove "Why".
q.cut(1)
return cleanup_statement(q)
try:
ModelSaver(saved_mdl,
mdl_name=mdl_name,
dir_save=model_path + mdl_name,
name_save='/saved_model_' + str(task))()
except:
print("Don't have saving model mode!")
log = dict()
log['y_pred'] = y_pred_
log['y_valid'] = y_valid_
save2pickle(log,
out_path=forecast_path + mdl_name,
out_file='/saved_model_' + str(task) + '.pickle')
if __name__ == "__main__":
from parser import get_parse
config = get_parse()
main(
mdl_name=config.mdl_name,
input_path=config.input_path,
model_path=config.model_path,
forecast_path=config.forecast_path,
batch_size=config.batch_size,
epochs=config.epochs,
task=config.num_jobs,
)
import json
import logging
import numpy as np
import copy
import torch
import torch.nn as nn
import random
import torch.optim as optim
from evaluator import OFAEvaluator
from torchprofile import profile_macs
from codebase.networks import NSGANetV2
from parser import get_parse
from eval_utils import get_dataset
args = get_parse()
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
device_list = [int(_) for _ in args.gpu.split(',')]
args.n_gpus = len(device_list)
args.device = torch.device("cuda:0")
if args.seed is None or args.seed < 0: args.seed = random.randint(1, 100000)
torch.cuda.manual_seed(args.seed)
torch.manual_seed(args.seed)
np.random.seed(args.seed)
random.seed(args.seed)
evaluator = OFAEvaluator(args,
model_path='../.torch/ofa_nets/ofa_mbv3_d234_e346_k357_w1.0')